INTERNATIONAL POLICY CENTER
Gerald R. Ford School of Public Policy
University of Michigan
IPC Working Paper Series Number 40
Women’s Working Status and Physical Spousal Violence in India
Yoo-Min Chin
April 2006
Women’s Working Status and Physical Spousal Violence in India
Yoo-Mi Chin
April 2006
Preliminary and Incomplete
Abstract
Empirical …ndings as well as theoretical predictions in the marriage bargaining literature
suggest that women’s …nancial independence has a positive e¤ect on their empowerment. Findings in the domestic violence literature, however, challenge the generalization of the results.
The theory of male backlash in the domestic violence literature predicts that in a patriarchal
economy, an increase in women’s economic independence will lead to an increase in cases of
domestic violence targeted at women. Violence is a means of restoring the husband’s authority
over his wife particularly when the women’s independence challenges the dominance of men.
Patterns of physical spousal violence in India are in line with the theory of male backlash in
a sense that working women are more subject to physical spousal violence than non-working
women. However, the interpretation is made di¢ cult by issues of reverse causality and omitted
variable bias. In this study, I address these issues by exploiting changes in rural women’s labor
market outcomes exogenously driven by the rainfall shocks and the rice-wheat dichotomy in
women’s employment. The IV regressions results indicate that women’s labor force participation decreases the probability of physical spousal violence by 0.07. The …ndings suggest that
the positive relationship between women’s working status and the physical spousal violence is
likely to be driven by reverse causality and omitted variable bias rather than the male backlash.
Department of Economics, Michigan State University, chinym@msu.edu I am grateful to Andrew Foster, Mark
Pitt, and Nancy Qian for their insight and guidance. I also thank Richard Blundell and Robert Pollak for their
comments and encouragement. Special thanks to Doug Park, Delia Furtado, Isaac Mbiti, and Muna Miky. All
remaining errors are mine.
1
1
Introduction
The theory of marital bargaining predicts that women’s greater …nancial independence should empower them with better outside options, lower their threshold for tolerating abuse inside marriage,
and lead to a reduction in violence against them. According to the male backlash theory, on the
other hand, in a patriarchal society, violence can be a means of restoring the husband’s authority
over his wife particularly when the women’s independence challenges the dominance of men. Therefore, an increase in women’s …nancial independence will increase the incidence of violence against
them.
Patterns of physical spousal violence in India are in line with the male backlash theory: women
who participate in the labor market tend to be more subject to physical violence by their husbands.
Given that women in India virtually do not have options outside marriage, which is an important
underlying assumption for the male backlash theory to be valid, the male backlash theory might
be a more appropriate model that captures the marital relationship in India.
However, the interpretation of such empirical …ndings is made di¢ cult by issues of endogeneity,
such as the reverse causality of women’s labor force participation as well as omitted variable bias.
For instance, the positive correlation between physical violence and women’s employment status
may re‡ect the causal e¤ect of domestic violence on the decision to work rather than the e¤ect
of working status on domestic violence. There might be a di¤erence between working women and
non-working women in terms of openness to public, which will lead to a systematic di¤erence in
reporting of violence between these two groups. In this paper, I address these issues by exploiting
the plausibly exogenous variation in rural women’s labor market outcomes driven by rainfall shocks
and the rice-wheat dichotomy in women’s employment and estimate the causal e¤ect of women’s
…nancial independence on the incidence of domestic violence. The IV regression results using the
interaction between the rainfall shocks and the rice-wheat dichotomy in women’s employment as
an instrumental variable indicates that women’s labor force participation decreases the probability
2
of physical spousal violence by 0.07. The results suggest that the positive relationship between
women’s working status and physical spousal violence is likely to be driven by endogeneity of
women’s working choice and omitted variable bias rather than the male backlash.
The paper proceeds as follows. In Section 2, the existing theories of domestic violence and
empirical …ndings are presented. Section 3 describes the conceptual framework. Section 4 describes
the data sets. Section 5 provides features of physical spousal violence and women’s attitudes towards
violence in India. The empirical speci…cations and estimation results are presented in Section 6,
and Section 7 concludes.
2
Background
2.1
2.1.1
Theoretical Background
Bargaining Theory of Domestic Violence
Noncooperative bargaining models of domestic violence predict that an increase in women’s economic independence will decrease the level of violence within the households. Women’s’…nancial
independence will increase their probability of leaving the relationship by providing favorable outside options, and lead to either the end of the relationship or a decrease in abusive treatment
within the intact households. Tauchen, Witte and Long (1991) developed a noncooperative model
of domestic violence where both a man’s and a woman’s utilities depend on domestic violence, the
behavior of woman, and his and her consumption of other goods. Both spouses can choose to make
an income transfer to each other and have threat point utilities, which are identical to utilities
outside the marriage. The e¤ect of changes in income depends on whether the threat point utility
is binding and whether there is a positive income transfer. When the woman’s threat point utility
is binding and there is a positive income transfer, an increase in the man’s income and an increases
in the woman’s income have opposite e¤ects on violence. As his income rises, the man can buy
more violence by increasing his …nancial transfer to her. As his payment for violence increases, the
woman’s tolerance of violence will also increase. As the woman’s income rises, the man is forced to
reduce the violence in order to maintain her reservation utility. When both individuals gain from
3
marriage and there are positive transfers, both persons’incomes have the same e¤ects on violence
and the e¤ects are in general negative.1 In a similar setting, Farmer and Tiefenthaler (1997)’s
noncooperative model of domestic violence also predicts that increase in a woman’s income will
decrease the level of violence because a woman’s …nancial independence increases her threat point.
2.1.2
Theory of Male Backlash
There are other models that generate opposite predictions to the theory of marital bargaining.
Those models characterized as theory of male backlash predict that women’s economic independence
could increase the physical spousal violence against them (Aizer 2005). Marital relationships are
governed by socially and culturally prescribed gender roles. To the extent that women’s economic
independence challenges socially sanctioned gender roles, women can be subject to more spousal
violence because the challenged man might try to reinstate his authority over his wife by in‡icting
violence on her (Macmillan and Gartner 1999). In this approach, women’s employment, for example
, does not merely provide an access to …nancial resources but also serves as a symbol that represents
the status of men and women within the households.
Similarly, according to the exchange theory (Molm 1990), a husband uses his ability to transfer
money and violence as the two sources of power. A husband can in‡uence his wife’s behavior by
transferring money to her or exercising violence as a punishment. As his wife’s income increases
relative to his, his ability to in‡uence his wife through monetary transfer will decrease, and he
will resort more to violence to in‡uence her behavior. Therefore, an increase in women’s …nancial
independence will lead to more spousal violence.
The models that focus on the symbolic nature of women’s economic independence are criticized
because they ignore women’s rationality constraint in abusive relationships (Aizer 2005). They do
not take into account the possibility that abused women can choose to end relationships. There
are certain cultures, however, in which women practically do not have outside options. In countries
where divorce or separation are accompanied by signi…cant stigma, the threat of ending the match
1
The direction of the e¤ect depends on how each person’s consumption of other goods a¤ect the man’s marginal
utility of violence. The assumption that his marginal utility of violence decreases with his consumption of other
goods does not necessarily rule out the positive e¤ect of income on violence. However, the violence can increase with
income only under very peculiar conditions.
4
may not be credible, in which case the bargaining model may not be appropriate (Luke and Munshi
2005).
2.2
Previous Empirical Findings
Empirical evidence on the e¤ect of economic independence of women on spousal violence is inconclusive. Using U.S. California county level data, Aizer (2005) examined the e¤ect of the relative
wage between female dominated sector (service) and male dominated sector (construction) on the
domestic violence. In her study, domestic violence rate was measured by arrests for domestic violence, female intimate partner homicides and hospitalizations for assault at a county level. She
found that increases in county level relative female wage over time decrease domestic violence at a
county level. Using 125 Californian women who were victims of domestic violence, Tauchen, Witte
and Long (1991) found that in low and middle income families, an increase in women’s income
reduces violence whereas an increase in men’s income increases violence. In high income families
where most of the income is earned by men, an increase in either party’s income will lower violence.
On the other hand, in high income families where most of the income is earned by women, an increase in her income will increase violence. Farmer and Tiefenthaler (1997) used victims of violence
data in the U.S. and found that higher female income leads to fewer incidence of violence. On
the other hand, increases in male earned income decreases violence, whereas increases in male unearned income increases violence. Macmillan and Gartner (1999) analyze the relationship between
women’s employment and spousal violence against them among Canadian women. Their empirical
results show that women’s employment increases risk of violence when husbands are unemployed,
whereas it decreases the risk when husbands are also employed.
The evidence in developing countries is more supportive of the male backlash theory. Luke and
Munsh (2005), for example, found out that controlling the total household income, an increase in
female income increases domestic violence against women among low caste families in Tamil Nadu
in India, which is likely to result from increase in disagreement over household resource allocations
as women’s …nancial independence increases. Bloch and Rao (2002) found that the risk of spousal
violence is higher for a woman from a rich household, using a survey data in three villages in
5
Karnataka in India. Their results suggest that a dissatis…ed husband whose cost of violence is low
enough will in‡ict violence on his wife in order to extract more monetary transfer from her family.
3
Conceptual Framework
Based on the previous literature, there can be two alternative hypotheses regarding the e¤ect
of women’s employment, representative of the female economic independence on the incidence of
spousal violence.
First, an increase in women’s labor force participation will decrease the incidence of spousal
violence. As the theory of marital bargaining suggests, an increase in female …nancial independence
will increase their probability of leaving the relationship by providing favorable outside options,
and lead to either the end of the relationship or a decrease in abusive treatment within the intact
households. Related to this argument, it has been suggested that increasing their job opportunities
in the labor market would be an e¤ective way to provide an outside option for women in developing
countries like India, where the lack of opportunities outside marriage for women is the major source
of unjust treatment of women before and within marriage. Bloch and Rao (2002), for example,
suggested that in India, "providing opportunities for women outside marriage and the marriage
market would signi…cantly improve their well-being by allowing them to leave an abusive husband,
by …nding a way of “bribing" him to stop the abuse, or by presenting a credible threat that achieves
the same objective. In more speci…c terms, the main opportunities for women outside the marriage
market would be in the labor market."
There is another theory in domestic violence literature that predicts a negative e¤ect of women’s
labor force participation on the incidence of violence. According to the exposure reduction theory,
when either husband or wife is working, spousal violence will decrease because they will have fewer
opportunities for con‡icts.
On the other hand, the alternative view suggests that an increase in women’s labor force participation will increase the incidence of spousal violence. This prediction might be con…ned to a
patriarchal society where social stigma against divorced or separate women is enormous and the
6
women’s threat of ending the relationships is incredible. In such a cultural surrounding, whenever
the women’s independence challenges the dominance of men, they might try to restore their authority by exercising more violence on their spouses. Similarly, as Bloch and Rao described in their
bargaining model, spousal violence can be a means to extract more transfer of resources. Therefore,
when divorce is tremendously costly and virtually not an option for women, a dissatis…ed husband
can exercise more violence on a woman from a richer family in order to extract more transfer from
her family. The same mechanism can be applied to working women who have more resources to be
extracted than non-working women.
On the other hand, a positive e¤ect of women’s labor force participation on the incidence of
violence might result from a totally di¤erent mechanism. For example, an increase in violence
reporting can be a labor market outcome. Therefore, it can be that we observe the positive
relationship not because working women experience more violence, but because working women
report more than non-working women do.
The purpose of this paper is to test these two opposite hypothesis. As will be shown in the
next section, patterns of physical spousal violence in India seem to be more supportive of the
male backlash theory in a sense that working women are subject to more physical spousal violence
than non-working women. However, the interpretation of such empirical …ndings is made di¢ cult
by issues of endogeneity. More speci…cally, the positive relationship can be a result of an omitted
variable bias. For example, labor force participation is positively correlated with poverty. At the
same time, poor women tend to be more subject to spousal violence because the lack of resources
serves as a stressor within the household. Therefore, it can be that violence is driven by the lack
of …nancial resources rather than the women’s working status. Similarly, there can be a systematic
di¤erence between working women and non-working women in reporting spousal violence. If working women tend to be more open to the public then non-working women, it is the di¤erence between
these two groups of women in terms of openness to public, not their working status.2 Moreover,
the results might be driven by the reverse causality. For instance, the positive correlation between
2
This argument is di¤erent from the argument that more reporting of violence is a labor market outcome. In this
argument, some unobservable characteristics of women are correlated with both labor market participation and the
reporting of violence, thereby biasing the e¤ect of working status. Therefore, an instrument will be a solution. If
more reporting is a labor market outcome, no instruments can …x the problem.
7
physical violence and women’s employment status may re‡ect the causal e¤ect of domestic violence
on the decision to work rather than the e¤ect of working status on domestic violence. Therefore,
I will address this endogeneity issues by exploiting the plausibly exogenous variation in women’s
labor market outcomes caused by rainfall shocks and the rice-wheat dichotomy in India and identify the causal relationship between women’s working status and the experience of physical spousal
violence.
4
Data
There are four sets of data employed in this study: the second National Family Health Survey
(NFHS-2) of India (1998-99), Indian District Database 1961-91, High Resolution Gridded Daily
Rainfall Data by the India Meteorological Department (IMD), Rural Economic and Demographic
Survey (1998-99).
The second National Family Health Survey (NFHS-2) of India was conducted between 1998
and 99. The NFHS-2 survey covers a representative sample of more than 90,000 eligible women
age 15–49 from 26 states that comprise more than 99 percent of India’s population. The survey
covers a variety of demographic and health issues including domestic violence, the main interest of
this paper. The data set contains information on women’s attitudes to domestic violence3 , everexperience of women’s domestic violence since age 15, persons who in‡icted violence, the incidence
of violence in the past 12 months, and frequency of the violence in the past 12 months. The survey
takes a single question approach. The respondent is asked a single question to determine whether
she has ever experienced violence. If she gives an a¢ rmative answer, then follow up questions are
asked. Given the sensitive nature of the issue, surveys dealing with domestic violence is particularly
subject to underreporting. In that sense, it is a shortcoming that women are given only one
chance to disclose their experience of violence. Moreover, violence is de…ned as “being mistreated
physically or beaten," and more concrete description of acts are not given. Because perceptions
about “physical mistreatment or beating" might vary by persons and household culture, it has
3
It is asked whether a husband is justi…ed in beating his wife in the following situations: if he suspects her of
being unfaithful, if her natal family does not give expected money, jewelry, or other items, if she shows disrespect for
in-laws, if she goes out without telling him, if she neglects the house or children, if she does not cook properly.
8
to be kept in mind that apart from the chronic underreporting issue, the unre…ned de…nition of
violence in NFHS-2 data might cause measurement problems.
Indian District Database provided state level crop area information. Since NFHS-2 provides
only state level identi…er, district level crop information in 1981 is aggregated at the state level.
Based on data availability, 18 states are chosen out of 26 states. These states divided into the rice
area and the wheat area depending on which crop is dominant in each state. The rice area includes
Andhra Pradesh, Assam, Bihar, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Manipur,
Meghalaya, Orissa, Tamil Nadu, and West Bengal. The wheat area includes Gujarat, Haryana,
Himachal Pradesh, Punjab, Rajasthan, and Uttar Pradesh. In each state, per capita total crop
area including 33 di¤erent crops4 is calculated.
High Resolution Gridded Daily Rainfall Data by the IMD was used to calculate the state level
rainfall shocks for the survey period. Rainfall shocks are measured as a deviation of the actual
rainfall in the past 12 months of the survey5 from the yearly normal (30 year average).
This study also used the Rural Economic and Demographic Survey 1998-99 in order to estimate the total labor income of the agricultural landless households. The total wage incomes of
the households are regressed on the basic demographic variables of the households and the crop
information of the state in which the households reside in. The parameter estimates are then used
to predict the total labor incomes of each household in the NFHS-2.
5
5.1
Violence in India
More Spousal Violence against Working Women
According to the National Family Health Survey 1998-9, working women are more likely to experience physical spousal violence than non-working women in India. Figure (1) presents the percentage
of married women6 who are beaten or physically since age 15 by the violence perpetrators. Out
4
Crops included are rice, jowar, bajra, maize, ragi, wheat, barley, gram, tur, groundnut, castor seed, sesamum,rapeseed/mustardseed, linseed, cotton, jute, msta, sugar, and tobacco
5
Among households within a same state, the actual rainfall in the past 12 months might be di¤erent depending
on in which months the survey was conducted.
6
Married women are de…ned as those women who are married and live with their husbands. Women who are
married but live separately from their husbands are excluded because separated women are di¤erent from women
9
of 80487 women who are married and live with their husbands, 35% women worked in the past
12 months and 65% women did not. Twenty seven percent of working women reported that they
experienced an act of physical violence perpetrated by somebody since age 15, whereas the corresponding …gure is 17% for non-working women. Twenty …ve percent of working women reported
that husbands were one of the perpetrators, and 15% of non-working women reported that their
husbands were one of the perpetrators. For 13% of non-working women, husbands were the only
people who ever beat or physically mistreated them, whereas 22% of working women reported that
their husbands were the only perpetrators of the violence. Compared to the ever experience of violence, the current violence rate is much lower. Figure (2) presents women’s experience of violence
in the past 12 months. In total, 12% of women reported that they experienced physical violence
by anybody in the last 12 months. Among working women, 14% experienced violence by anybody,
whereas 10% of non-working women experienced violence. Husbands are again main perpetrators
of the physical violence against women. Nine percent of all women reported that husbands were
only perpetrators of violence in the past year. Among working women, 12% said that husbands
were the only perpetrators whereas the corresponding …gure for non-working women was 7 percent.
Therefore, domestic violence against women is predominantly committed by their intimate
partners, and more physical violence is in‡icted on working women than non-working women.
If the latter re‡ects the causal relationship - e¤ect of women’s labor force participation on the
incidence of violence -, the patterns of physical spousal violence in India seem to be in line with
the theory of male backlash.
5.2
Women’s Attitudes towards Spousal Violence
Table (2) presents women’s attitudes towards spousal violence. As mentioned earlier, NFHS-2 asked
women whether wife beating is justi…ed in the following situations: if a husband suspects that his
wife is being unfaithful, if her natal family does not give expected money, jewelry, or other items,
if she shows disrespect for in-laws, if she goes out without telling him, if she neglects the house
or children, and if she does not cook properly. Overall, 32 percent of women agree with beating
who live with their husbands in terms of exposure to the risk of violence.
10
as a punishment for being unfaithful. Relatively few women agree with violence as a punishment
for insu¢ cient dowry. Seven percent of women believe that a wife deserves beating if there was
not enough dowry or monetary transfer from the wife’s family to the husband’s. More than thirty
percent of women agree with physical punishment if a wife shows disrespect for in-laws, or if she
goes out without telling her husband. In general, taking care of house and children are considered as
the most important responsibility of women. About 40 percent of women believe that they deserve
beating if they neglect these duties. Further, around 24% of women agree with beating if they do
not cook properly. Columns (2) and (3) present women’s attitudes towards spousal violence by
their working status. Notably, working women are more likely to accept violence as a punishment
than non-working women, in all the occasions.
6
Empirical Analysis
6.1
Identi…cation and Sample Selection
The main interest of this paper lies in understanding the causal relationship between a woman’s
working status and the incidence of physical spousal violence. Since I only have information on
women’s working status in the past 12 months, I will focus on the e¤ect of a woman’s working status
in the past 12 months on her experience of violence in the past 12 months. Therefore, caution is
required in interpreting the results because short term variations in women’s working status might
have rather restricted implications for the experience of spousal violence.
The major concern in examining the relationship between women’s working status and their
experience of violence is that women’s working status is endogenous. A variable that simultaneously
a¤ects both her working choice and incidence of violence might bias the results of standard linear
probability estimation. For example, poverty can cause higher incidence of violence as well as
higher labor force participation of women. If an extroverted woman not only tends to choose to
participate more in the labor market but also is more likely to report experience of spousal violence,
the coe¢ cient of a women’s working status will be biased upward. Moreover, a woman who su¤ers
more from spousal violence can choose to work more outside home if marginal disutility of working
11
decreases with level of violence. Therefore, in order to address this endogeneity of women’s working
status, I need some exogenous factors that change women’s working status but are uncorrelated
with unobservable violence factors within households or women’s openness to public.
When the sample is restricted so that it only includes agricultural landless households, a rainfall
shock might be a valid instrument. It is because a rainfall shock will exogenously change women’s
working status, but it is unlikely to be correlated with unobservable violence factors within households or women’s openness to public. However, a major concern in using rainfall shocks as an
instrument is that rainfall shocks might violate the exclusion restriction through other channels
such as the husband’s labor incomes.
Another source of exogenous variations in women’s working status might be the rice-wheat
dichotomy in India. In India. female labor force participation rates are consistently lower in the
traditional wheat-growing belt of the northwest than in the rice-growing eastern and southern states.
This geographically distinct employment variation is related to di¤erences in farming intensities
and cropping patterns across regions: in the wheat-growing region where plough cultivation is
predominant, demand for female labor is low, whereas in the rice-growing region where weeding
and transplanting is prevalent, demand for female labor is high (Boserup 1970). A number of
studies analyzed these variations in female labor employment generated by di¤erential demand for
female labor due to ecological variations in cropping patterns and how these economic values of
women a¤ect household decision making (Bardhan, 1984; Rosenzweig and Schultz, 1982; Miller,
1981). For example, Bardhan relates North-South di¤erence in survival chances of female child to
the rice-wheat dichotomy and the resulting di¤erential patterns of female employment in India. On
the other hand, we do not observe this distinct pattern for men, because men engage in both types
of work, whereas women are largely excluded from ploughing. Therefore, whether the households
reside in a state where more rice is grown than wheat might be an exogenous factor that a¤ects
women’s working status. However, if a more patriarchal and violence-oriented family tends to choose
either area for any reason, being in a rice or wheat state might be correlated with unobservables
in the violence equation Table (3), for example, presents women’s attitudes towards violence by
crop states. Other than for being unfaithful, for all the other occasions, women in the rice state
12
are more likely to accept violence as a punishment mechanism. If this re‡ects di¤erent household
culture by crop state, it is likely that those violent oriented households choose more to reside in the
rice state or households in the rice state become more patriarchal due to some social surroundings
speci…c to that region.
7
Rainfall shocks are likely to violate the exclusion restriction because they will a¤ect the physical
spousal violence through other channels such as husbands’labor income. Being in a rice or wheat
state might be correlated with unobservables in the main equation if a more patriarchal and violenceoriented family tends to choose either state. However, the interaction between the rainfall shocks
and the rice state dummy will not be correlated with the husband’s income because we do not
observe rice-wheat dichotomy for men’s employment. The main idea is that labor demand shocks
created by rainfall shocks will be di¤erential for women’s labor force participation depending on
whether she is in a rice area or in a wheat area, whereas for men who engage in both rice and wheat
production no di¤erential e¤ects are expected. Further, the shocks are not likely to be correlated
with the household’s intrinsic orientation for violence, because it is hard to imagine that unexpected
weather shocks can change personality or household culture other than through income changes
and vice versa. Therefore, the interaction between the dummy for the rice area and the rainfall
shocks will be a valid instrument in identifying the e¤ect of women’s labor force participation on
the physical spousal violence.
There are several concerns in using the interaction term as the instrument. First, rainfall might
di¤erentially a¤ect the demand for male labor in both areas as well, if one of the crops is more
sensitive to rain. Since rice production in general is more dependent on water availability, it is
likely that the e¤ect of rainfall on men’s labor income will be also di¤erential in the two areas.
This might compound the result of the IV regression. I address this issue by directly controlling
predicted total agricultural household income.8 Second, rainfall shock might have a direct e¤ect
7
This is contrary to the general belief that North is more conservative and oppressive in terms of treatment
of women than the South. As far as the spousal violence is concerned, it seems that the rice area which largely
corresponds to the South is more patriarchal.
8
Using the Rural Economic and Demographic Survey 1998-99, I estimated landless househould total labor income
equation. The estimation is based on a set of household level demographic variables and a set of state level agricultrual
variables. Speci…ally, the household level demographic variables include the number of men, the number of women,
the mean age of men, the squared mean age of men, the squared mean age of women, the mean education of men,
the mean education of women, the squared mean education of men, the squared education of women. The state level
13
on violence. It is unlikely that the household culture or its violence orientation is a function of a
rainfall shock. However, rainfall shocks might change the household time allocation patterns. If,
for example, more rainfall shocks cause the couple to spend more time within the household than
outside, the risk of violence might increase. If households in one area have a more violent and
patricarchal culture than the other, the risk of violence caused by rainfall shocks will be higher in
one area than in the other. Then, there will be a correlation between the interaction term and the
unobservables in the main equation. However, Table (3) suggests that the households in the rice
area are likely to have a more violent culture than the households in the wheat area, if at all. If
a rainfall shock in the rice area increases both women’s labor force participation and the risk of
violence more than those of women in the wheat area, the IV regression results will be upwardly
biased. Therefore, if the sign of coe¢ cient of the women’s working status is negative even with the
bias, the bias cannot qualitatively change the conclusion.
6.2
Estimation Equations
The second stage equation is de…ned by
Vi =
0
+
1 Wi
+ Xi + rs + Sts + "i
(1)
where V is the woman’s violence experience in the past 12 months9 , W is the woman’s working
status in the past 12 months, X include household demographic variables, wealth (assets), household
labor income, r is the dummy for being in a rice state, Sts is the rainfall shock that varies by state
and the survey month.
aricultural variables include per capita total crop area, a dummy indicating whether it is (majorly) a rice growing
state, rainfall shock, rainfall shocks interacted with other two crop variables. The estimated coe¢ cients were used to
predict the total household labor income of the sample households in the original data set (Demographic and Health
Survey).
9
The experience of violence is measured in two di¤erent ways. First it is a dummy variable that takes 1 if physical
spousal violence towards woman took place in the past 12 months and takes 0 otherwise. Violence is also measured
in terms of frequency. The variable is measured as a discrete variable that takes 0 if no violence was in‡icted, 1 if it
took place once, 2 if it happened a few times, and 4 if violence was perpetrated many times.
14
The …rst stage equation is de…ned by
Wi =
where Sts
0
+
1 [Sts
rs ] + Xi + rs + Sts + "i
(2)
rs is the interaction between the rainfall shock and the rice state dummy.
For estimation, two di¤erent speci…cations are employed. First, I used the interaction term as
the only instrument. Therefore, the main e¤ects of the rainfall shock and the rice state dummy
were included in the second stage equation. As additional controls in the second stage, I included
state level per capita total crop area, village level arable land area, village level distance from the
nearest town. These variables are included because they might a¤ect violence through women’s
working status as well as the household labor income. Since I do not have the household income, I
directly control these variables instead.
In the second speci…cation, I directly control the predicted household labor income. As explained
earlier, I predicted the household labor income based on their demographic variables, the crop
variables both at the state and at the village level, distance from the nearest town, rice state
dummy, rainfall shocks, and the interactions between the rainfall shock and the crop variables.
Therefore, any e¤ects of these crop variables and the rainfall shocks through the household labor
income are controlled. Since it is unlikely that these crop variables will directly a¤ect the spousal
violence other than through the household income, once their e¤ects through the total household
income are controlled, I used these variables as additional instruments. In doing so, the main
e¤ects of the rainfall shocks and the rice state dummy are also treated as excluded instruments.
The overidenti…cation test results support the validity of the second speci…cation.
6.3
6.3.1
Results
Estimation with Entire Sample
Table (8) presents the estimation results of equation (1) using the entire sample of women. The
summary statistics of the entire sample is reported in Table (4). All the columns are results of linear
probability model estimations. Column (1) presents that women’s labor force participation has a
15
signi…cant positive e¤ect on the probability of physical spousal violence. When a woman works, the
probability of physical spousal violence increases by 0.04. The coe¢ cient decreases approximately
by half when wealth10 is also controlled, suggesting that positive correlation between poverty and
women’s working status tends to bias the coe¢ cient upward. Inclusion of other demographic
variables decreases the coe¢ cient slightly more but it is still positive and signi…cant.
Women’s age decreases the probability of spousal violence. A one year increase in women’s age
will decrease the probability of violence by 0.0017. On the other hand, men’s age has a positive
but insigni…cant e¤ect on the physical spousal violence. This result suggests that the age di¤erence
re‡ects their relative status in the marital relationship and older women bene…t from the better
status. The more children they have, the more violence they experience. This may be driven by
the fact that the available resource per capita decreases with family size and this lack of resources
generates more stress within the household. Both women’s and men’s education have a signi…cant
negative e¤ect on the probability of spousal violence. A one year increase in women’s education
leads to decrease in the probability of physical spousal violence by 0.0018, whereas a one year
increase in husbands’education decreases the probability by 0.002. Women residing in urban areas
are signi…cantly more likely to su¤er from spousal violence. When the couple lives in urban areas,
the probability of spousal violence increases by 0.025. In general, urban households tend to have
less contacts with neighbors and communities than rural households. Violence can be more easily
committed when issues within the household are less exposed to public attention (Kishor and
Johnson 2004). Further, being a low caste increases the probability of spousal violence by 0.02.
In Table (8) columns (4)-(6), the e¤ect of women’s labor force participation on the frequency of
violence in the past 12 months exhibit similar patterns. The frequency of violence is measured as a
discrete variable which takes 0 if no violence was in‡icted, 1 if it took place once, 2 if it took place a
few times, and 4 if violence was perpetrated many times. Women’s labor force participation has a
positive and signi…cant e¤ect on the frequency of violence after controlling demographic variables as
well as wealth. Women’s age as well as both men’s and women’s education decrease the frequency
10
The wealth index was constructed using household asset data and principal components analysis. Assets include
a number of consumer items such as a telephone, bicycle or car as well as availability of drinking water and sanitation
facilities and etc. Each asset is assigned a score generated through principal components analysis and the scores are
summed up by household. (Kishor and Johnson 2004)
16
of violence, whereas urban living and low caste increase it. However, the number of children does
not have a signi…cant e¤ect on the frequency of violence.
6.3.2
IV Estimation with Restricted Sample
Reduced form and First-Stage Results
Table (9) presents the reduced form e¤ect of rainfall
shock on the incidence of spousal violence in the rice state and in the wheat state respectively.
This is the results using only the agricultural landless household data. The summary statistics of
the restricted sample is reported in Table (5). Columns (1) and (3) present the e¤ects of rainfall
shocks on the incidence of violence when the shock is the only control variable. In the rice area,
one more mm of rainfall shock decreases the incidence of violence by 0.013 percentage points, when
the rainfall shock is the only control. The precision declines as the other controls are included
and rainfall e¤ects become signi…cant at 10% level, if all the included instruments other than the
total household labor income are controlled (Column (2)). One more mm of the rainfall shock will
decrease the probability of the incidence of violence by 0.00011. The e¤ects of rainfall shocks are
largely insigni…cant in the wheat area when only the rainfall shock is controlled (Columns (3)).
When all the exogenous variables are controlled, the rainfall shocks have positive e¤ects of the
incidence of violence and the e¤ect is signi…cant at 10% level (Column (4)). One more mm of
rainfall shock increases the incidence of violence in the wheat area by 0.024 percentage points.
Table (10) presents the reduced form e¤ects of rainfall shock on the frequency of violence in the
rice area and in the wheat area respectively. The result is not qualitatively di¤erent from Table
(5).
Table (12) presents the …rst stage regression results. Column (1) presents the e¤ect of the
interaction between the rice state dummy and the rainfall shock on the probability of the women’s
working, when the household labor income is not controlled. When there is one more mm of rainfall
and the woman is in the rice state, her probability of working will increase by 0.002. The results
suggest that compared to women in the wheat area, the rainfall shocks a¤ect women in the rice
area more favorably, which is in line with the rice-wheat dichotomy. Column (2) presents the
…rst stage regression results when the imputed household labor income is directly controlled. In
17
column (2), all the crop variables, the rainfall shocks, and their interactions are being treated as
excluded instruments. The e¤ect of the interaction between the rainfall shock and the rice state
dummy is still signi…cantly positive when the household income is controlled. Notably, the main
e¤ect of the rainfall shock on women’s working status is signi…cantly negative. One more mm of
rainfall shock will decrease women’s labor force participation by 0.2 percentage points. The e¤ects
of rainfall shock interacted with the per capita crop area are signi…cantly positive. One more mm of
rainfall interacted with one more hectare per capita crop area will increase women’s working by 0.2
percentage points. Both the state level per capita crop area and the village level arable land have
positive and signi…cant e¤ects on women’s working status. Being in the rice state has a positive
but largely insigni…cant e¤ect on the working status. All the instruments are jointly signi…cant.
Women’s Working Status and the Physical Spousal Violence
Table (13) presents the
e¤ect of women’s labor force participation on the incidence of physical spousal violence. The OLS
results in column (1) show that when women work, the probability of physical spousal violence will
increase by 0.06. Again, wealth has a signi…cant negative e¤ect on the incidence of violence and the
inclusion of wealth decreases the e¤ect of the working status, suggesting the positive correlation
between poverty and women’s labor force participation. The coe¢ cient becomes a little bit smaller
when other demographic controls are included (column (3)). Among the demographic variables,
only the women’s education and the number of children have signi…cant e¤ects. A one year increase
in women’s education will decrease the probability of violence by 0.003. One more child will increase
the probability of violence by 0.006. However, as is the results with the entire sample, even after
controlling the e¤ect of wealth, women’s labor force participation still increases the probability of
spousal violence by 0.04.
Column (4) presents the IV regression results using the interaction between the rice state
dummy and the rainfall shock as the only instrument. Once instrumented, the women’s working
status has a negative e¤ect on the incidence of violence. However, the e¤ect is not signi…cant at
the conventional level and only signi…cant at 15% level. Column (5) presents the IV regression
results when the imputed household labor income is directly controlled and all the crop variables
18
and rainfall shocks are treated as excluded instruments.11 As is apparent in the chi square p-value
for the overidenti…cation test, the full set of instruments passes the overidenti…cation test. The
results in column (6) show that the exogenous changes in women’s working status decreases the
probability of physical spousal violence by 0.07, and the e¤ect is signi…cant at the conventional
level. The results suggest that the positive e¤ect of women’s labor force participation on the
incidence of violence in the OLS regression is likely to be driven by reverse causality or omitted
variable bias rather than the male backlash. Among the demographic variables women’s education
has a signi…cant negative e¤ect on the probability of violence. Although men’s education also has
a negative e¤ect on the incidence of violence, the e¤ect is not statistically signi…cant. Being a
low caste increases the probability of violence by 0.04, whereas wealth decreases the probability
of violence. The e¤ect of the total household income on the incidence of violence is negative but
statistically not signi…cant.
Table (14) presents the e¤ect of women’s working status on the frequency of violence. The
results are qualitatively similar with Table (13). The simple OLS results suggest that women’s
labor force participation increases the frequency of violence even after controlling demographic and
wealth variables. However, when the working status is instrumented, it has a signi…cant negative
e¤ect on the frequency of violence (column (5)). Among the demographic variables, both women’s
and men’s education signi…cantly decrease the frequency of violence. Again, wealth decreases the
frequency whereas being a low caste increases the frequency. An increase in the total household
labor income signi…cantly decreases the frequency of violence.
Women’s Income Contribution and the Physical of Violence
Table (15) presents the
relationship between the degree of women’s contribution to household income and the incidence
of physical spousal violence. Women’s contribution to the household income is a discrete variable
that takes 0 if women do not work, 1 if the contribution is almost none, 2 if it is less than half, 3
if it is about half, 4 if it is more than half, 5 if it is all. The OLS results show similar patterns as
11
The instruments are the rainfall shocks, rice state dummy, the interaction between the rice state dummy and the
rainfall shock, state level per capita crop area, interaction between the per capita crop area and the rainfall shocks,
village level per capita arable land, village level distance from the nearest town.
19
the working status results. More contribution leads to more violence and the relationship is robust
to the inclusion of other demographic variables and the wealth (columns (1)-(3)).
The …rst stage regression result for women’s income contribution using the same set of instruments for women’s working status is reported in Table (12). As for the working status, the
interaction between the rice state dummy and the rainfall shock has a signi…cant and positive e¤ect
on the woman’s contribution to the total household income (columns (3) and (4)). In column (4),
the additional instruments are added and the total household labor income is controlled. The main
e¤ect of the rainfall shock is negative. Its e¤ects through the per capita total crop area is positive
but signi…cant at 10% level. Having more crop areas at a state level and more arable land at a
village level increase women’s contribution to the household income. Being in a rice state increases
women’s contribution to the household income, but the e¤ect is not statistically signi…cant. The
instruments are jointly signi…cant
Column (4) and (5) in Table (15) presents IV regression results. Exogenous increases in women’s
contribution to the household income decrease the incidence of spousal violence. However, the e¤ect
is not signi…cant at the conventional level. Column (5) presents the results controlling the predicted
household income and including additional instruments. The increase in women’s contribution to
the household income decreases the probability of violence, but the e¤ects are not statistically
signi…cant even at 10% level. Among the demographic variables, women’s education and wealth
have signi…cant negative e¤ects, whereas being in a low caste signi…cantly increases the incidence
of violence. All the other demographic variables are not statistically signi…cant at the conventional
level. Table (16) presents the e¤ect of women’s income contribution on the frequency of violence.
The results are similar to the results in Table (15). The simple OLS results suggest that women
who contribute more to the total household income will su¤er more from physical spousal violence.
The IV results suggest the opposite, although the e¤ect is signi…cant only at 10 % (column (5)).
20
7
Conclusion
The purpose of this study is to identify the causal relationship between women’s working status
and the risk of spousal violence against them. In India, working women tend to be more subject to
physical spousal violence than non-working women. Given that there is virtually no option outside
marriage for women in India, the theory of male backlash seems to appropriately explain reasons
for more spousal violence against working women in India. However, there are concerns that the
positive relationship between women’s labor force participation and physical spousal violence might
be driven by reverse causality or omitted variable bias. In this paper, I address these issues by
exploiting plausibly exogenous variations in rural women’s working status driven by rainfall shocks
and the rice wheat dichotomy. The IV regression results indicate that women’s working status has a
signi…cant negative e¤ect on the incidence and the frequency of physical spousal violence. Women’s
labor force participation will decrease the probability of physical spousal violence by 0.07. The
frequency of violence will decrease by 0.21 if a woman works. Therefore, the positive relationship
between women’s working status and the experience of violence in the simple linear probability
model seems to be driven by endogeneity of women’s labor force participation rather than the male
backlash. These results suggest that increasing women’s human capital and expanding their job
opportunities in the labor market will improve their well-being by increasing their status within
the household, as well as by decreasing the risk of physical spousal violence against them.
21
References
[1] Aizer, Anna. "Wages, Violence, and Health in the Household." mimeo Brown University
(2005).
[2] Asling-Monemi, Kajsa., Rodolfo Pena, Mary C. Ellsberg, and Lars A. Persson. “Violence
against women increases the risk of infant and child mortality: A case-referent study in
Nicaragua.” Bulletin of the World Health Organization Vol. 81, No.1 (2003), 10-18.
[3] Bawah, Ayaga.A., Patricia Akweongo, Ruth Simmons, and James F. Phillips. “Women’s fears
and men’s anxieties: The impact of family planning on gender relations in northern Ghana.”
Studies in Family Planning, Vol. 30, No. 1 (1999), 54-66.
[4] Becker, Gary S. “A theory of marriage: Part I.” Journal of Political Economy, Vol. 81, No. 4
(1973), 813-846.
[5] Bloch, Francis and Vijayendra Rao. “Terror as a Bargaining Instrument: A Case Study of
Dowry Violence in Rural India.”American Economic Review. Vol 92, No. 4 (2002), 1029-1043.
[6] Byrne, Christina A., Heidi S. Resnick, Dean.G. Kilpatrick, Connie L. Best, and Benjamin
E. Saunders. “The socioeconomic impact of interpersonal violence on women.” Journal of
Consulting and Clinical Psychology, Vol. 67, No. 3 (1999), 362-366.
[7] Dyson, Tim, and Mick Moore. “On kinship structure, female autonomy and demographic
behavior in India.” Population and Development Review, Vol. 9, No. 1 (1983), 35-60.
[8] Farmer, Amy and Jill Tiefenthaler. “Domestic Violence: The Value of Services as Signals.”
American Economic Review, Vol. 86, No. 2 (1996), 274-279.
[9] Farmer, Amy and Jill Tiefenthaler. “An Economic Analysis of Domestic Violence.” Review of
Social Economy, Vol. 55, No. 3 (1997), 337-358.
[10] Foster, Andrew D. "Marriage-Market Selection and Human Capital Allocations in Rural
Bangladesh." manuscript (2002).
22
[11] Garcia, Brigida. (ed.). Women, poverty and demographic change. Liege: Oxford University
Press for IUSSP (2000).
[12] Goetz, Anne M.“Managing organisational change: The “gendered” organisation of space and
time.” Gender and Development, Vol.5, No. 1 (1997), 17-2.
[13] Goetz, Anne M., and Rina Sen Gupta. “Who takes the credit? Gender, power and control over
loan use in rural credit programs in Bangladesh.”World Development, Vol. 24, No. 1 (1996.),
45-63.
[14] Heise, Lori L. “Reproductive freedom and violence against women: Where are the intersections?” Journal of Law, Medicine and Ethics, Vol. 21, No. 2 (1993), 206-216.
[15] Heise, Lori, Jacqueline Pitanguy, and Adrienne Germain. “Violence against women: The hidden health burden.”World Bank Discussion Paper #225. Washington D.C.: The World Bank
(1994).
[16] Hornung, Carton A., B. Claire McCullough, and Taichi Sugimoto. “Status relationships in
marriage: Risk factors in spouse abuse.” Journal of Marriage and the Family, Vol. 43, No. 3
(1981), 675-692.
[17] Kishor, Sunita and Kiersten Johnson. "Pro…ling Domestic Violence- A Multi-Country Study."
Publication (OD31) Measure DHS+, (2004).
[18] Macmillan, Ross and Rosemary Gartner. “When She Brings Home the Bacon:Labor Force
Participation and the Risk of Spousal Violence Against Women.”Journal of Marriage and the
Family, Vol. 61, No.4 (1999), 947-958.
[19] Malhotra, Anju , and Mark Mather. “Do schooling and work empower women in developing
countries? The case of Sri Lanka.” Sociological Forum, Vol.12, No. 4 (1997), 599-630.
[20] Martin, Sandra L., Amy O. Tsui, Kuhu Maitra, and Ruth Marinshaw. “Domestic violence in
northern India.” American Journal of Epidemiology, Vol.150, No.4 (1999), 417-426.
23
[21] Mueller, Charles W., Toby L. Parcel, and Fred C. Pampel. “The e¤ect of marital dyad status
inconsistency on women’s support for equal rights.” Journal of Marriage and the Family,
Vol.41. No.4 (1979), 779-791.
[22] National Academy Press. Growing Populations, Changing Landscapes: Studies from India,
China, and the United States, Washington, D.C.: National Academy Press (2001).
[23] Luke Nancy and Kaivan Munshi. “Women as Agents of Change: Female Income, Social A¢ liation and Household Decisions in South India.”, mimeo Brown University (2005).
[24] Pollak, Robert A. “ Intergenerational Model of Domestic Violence.” Journal of Population
Economics, Vol.17 (2004), 311-329.
[25] Pollak, Robert A and Shelley Lundberg. “Noncooperative Bargaining Models of Marriage.”
American Economic Review, Vol.84, No. 2 (1994), 132-137.
[26] Pollak, Robert A and Shelley Lundberg. “Separate Spheres Bargaining and the Marriage Market”, Journal of Political Economy, Vol.101, No.6 (1994), 988-1010.
[27] Senauer, Benjamin, Marito Garcia and Elizabeth Jacinto. "Determinants of the Intrahousehold
Allocation of Food in the Rural Philippines." American Journal of Agricultural Economics,
Vol.70, No.1 (February 1988), 170-180.
[28] Tauchen, Helen V., Witte, Ann Dryden, and Sharon K. Long, “Violence in the Family: A
Non-random A¤air.” International Economic Review, Vol.32, No. 2, (May 1991), 491-511.
[29] Vanneman,
1961-1991:
Reeve
and
Douglas
machine-readable
Barnes.
data
…le
and
2000.
Indian
codebook.
District
Internet
http://www.inform.umd.edu/~districts/index.html. College Park, Maryland:
Data,
address:
Center on
Population, Gender, and Social Inequality.
[30] Yick, Alice G. “Feminist theory and status inconsistency theory: Application to domestic
violence in Chinese immigrant families.” Violence Against Women, Vol.7, No. 5 (2001), 545562.
24
0.3
0.25
0.2
Work
No Work
0.15
0.1
0.05
0
Anybody
Husband
Husband only
Figure 1: Violence Experience (ever) by Working Status and Perpetrator
25
0.16
0.14
0.12
0.1
Work
No Work
0.08
0.06
0.04
0.02
0
Anybody
Husband
Husband only
Figure 2: Violence Experience (past year) by Working Status and Perpetrator
26
Total
Work
No Work
Violence ever
Anybody
Husband
Husband only
0.21
0.19
0.16
0.27
0.25
0.22
0.17
0.15
0.13
0.12
0.11
0.09
0.14
0.14
0.12
0.1
0.09
0.07
80487
28860
51627
Violence past 12 months
Anybody
Husband
Husband only
Observation
Table 1: Violence Experience by Working Status and Perpetrator
27
Total
Work
No Work
Husband may hit wife
1) if she is unfaithful
Yes
No
0.32
0.67
0.37
0.62
0.3
0.69
2) if her family does not give money
Yes
No
0.07
0.93
0.1
0.89
0.05
0.95
3) if she shows disrespect
Yes
No
0.34
0.66
0.41
0.59
0.3
0.7
4) if she goes out without telling him
Yes
No
0.36
0.63
0.44
0.55
0.31
0.68
5) if she neglects house or children
Yes
No
0.4
0.6
0.49
0.51
0.34
0.66
6) if she does not cook properly
Yes
No
0.24
0.75
0.31
0.68
0.2
0.8
75884
26871
49013
Observation
Table 2: Attitudes towards Violence by Working Status
28
Total
Rice
Wheat
Husband may hit wife
1) if she is unfaithful
Yes
No
0.33
0.66
0.32
0.67
0.38
0.61
2) if her family does not give money
Yes
No
0.07
0.92
0.08
0.91
0.04
0.96
3) if she shows disrespect
Yes
No
0.35
0.64
0.39
0.6
0.23
0.76
4) if she goes out without telling him
Yes
No
0.4
0.6
0.42
0.57
0.29
0.71
5) if she neglects house or children
Yes
No
0.42
0.57
0.47
0.52
0.26
0.73
6) if she does not cook properly
Yes
No
0.25
0.75
0.26
0.73
0.19
0.81
19074
13742
5323
Observation
Table 3: Attitudes toward Violence by Crop State
Observation
Mean
Mean (weighted) Standard deviation
Violence in the past year
75884
0.08
0.1
0.28
Work in the past year
75884
0.35
0.38
0.48
Woman age
75884
31.17
30.8
8.64
Man age
75884
37.12
36.95
9.83
Number of children
75884
2.65
2.63
1.78
Woman education
75884
3.96
3.62
4.79
Man education
75884
6.56
6.23
5.1
Urban
75884
0.32
0.27
0.47
Low caste
75884
0.58
0.6
0.49
Wealth
75884
0.03
-0.12
1
Table 4: Summary Statistics- Entire Sample
29
Observation
19074
Mean
0.12
Mean (weighted)
0.13
Standard deviation
0.33
Work in the past year
19074
0.39
0.42
0.49
Woman age
19074
30
29.84
8.91
Man age
19074
36.34
36.34
10.2
Number of children
19074
2.62
2.58
1.84
Woman education
19074
2.36
2.29
3.68
Man education
19074
4.34
4.22
4.44
Low caste
19074
0.72
0.71
0.45
Wealth
19074
-0.46
-0.52
0.72
Household labor income
19704
11764
12021
12745
Rainfall shock
19074
104.65
142.54
188.2
Total crop area (per capita)
19074
0.3
0.28
0.13
Arable land (per capita)
19074
19.68
19.43
582.45
Distance
19074
32.84
43
433.6
Violence in the past year
Table 5: Summary Statistics-Agricultural Landless Households
30
Observation
13751
Mean
0.14
Mean (weighted)
0.14
Standard deviation
0.34
Work in the past year
13751
0.44
0.46
0.5
Woman age
13751
29.98
29.89
8.95
Man age
13751
36.87
36.82
10.26
Number of children
13751
2.54
2.48
1.8
Woman education
13751
2.5
2.45
3.74
Man education
13751
4.14
4.08
4.36
Low caste
13751
0.72
0.71
0.45
Wealth
13751
-0.57
-0.56
0.66
Household labor income
13751
10073
10973.54
13230.61
Rainfall shock
13751
83.2
143.62
205.6
Total crop area (per capita)
13751
0.26
0.26
0.12
Arable land (per capita)
13751
3.73
4.72
74.85
Distance
13751
39.76
50.22
496.07
Violence in the past year
Table 6: Summary Statistics - Households in Rice Area
31
Observation
5323
Mean
0.09
Mean (weighted)
0.1
Standard deviation
0.28
Work in the past year
5323
0.25
0.27
0.43
Woman age
5323
30.03
29.69
8.8
Man age
5323
34.96
34.61
9.9
Number of children
5323
2.84
2.89
1.9
Woman education
5323
1.99
1.71
3.5
Man education
5323
4.86
4.73
4.61
Low caste
5323
0.73
0.71
0.45
Wealth
5323
-0.18
-0.33
0.77
Household labor income
5323
16132.05
15730.73
10172.38
Rainfall shock
5323
160.08
138.7
116.03
Total crop area (per capita)
5323
0.4
0.37
0.98
Arable land (per capita)
5323
60.88
71.44
1094.96
Distance
5323
14.95
17.49
193.81
Violence in the past year
Table 7: Summary Statistics - Households in Wheat Area
32
Violence incidence
Violence frequency
(1)
(2)
(3)
0.044538 0.021215 0.019399
(5.19)**
(2.96)**
(3.43)**
Woman age
-0.001693
(3.88)**
Man age
0.000285
-0.56
Number of children
0.004393
(2.29)*
Woman education
-0.001819
(2.41)*
Man education
-0.002075
(4.42)**
Urban
0.025133
(5.56)**
Low caste
0.019284
(2.28)*
Wealth
-0.044911 -0.033968
(8.37)**
(7.38)**
Constant
0.080997 0.084215 0.117483
(7.93)**
(11.92)** (8.49)**
Working status
Observations
75884
75884
75884
(4)
(5)
(6)
0.094466 0.045443 0.040229
(5.12)**
(2.88)**
(3.12)**
-0.002831
(2.87)**
0.000618
-0.55
0.006904
-1.76
-0.003283
(2.14)*
-0.00499
(6.53)**
0.056719
(6.15)**
0.037729
(2.38)*
-0.0944 -0.074227
(8.39)**
(7.33)**
0.157033 0.163798 0.219515
(7.79)** (12.36)**
(8.47)**
75884
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Table 8: OLS Estimation Results - Entire Sample
33
75884
75884
34
Violence incidence
(3)
(4)
-0.00028049 0.00024245
-0.98
-2
-0.7070833
(6.22)**
-1.176E-05
(8.99)**
0.00001798
(5.03)**
0.24203507 0.48342151
(2.81)*
(9.39)**
Violence incidence
(1)
(2)
-0.00013 -0.00011442
(2.45)*
-1.98
-0.15623488
-2.04
0.00002117
-0.7
-0.00000115
-0.19
0.155691 0.19306874
(7.56)**
(5.73)**
Table 9: Reduced Form Estimation by Crop State - Violence Incidence
Observations
13751
13751
5323
5323
Columns (1)-(4): Linear Probability Model
Colmuns (1) and (3) : Controls - rainfall shock
Columns (2) and (4): Controls - all excluded as well as included instruments other than the
total household income
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Constant
Distance
Arable land (per capita)
Total crop area (per capita)
Rainfall shock
Wheat major state
Rice major state
35
Violence frequency
(3)
(4)
-0.00028049 0.00024245
-0.98
-2
-0.70708331
(6.22)**
-0.00001176
(8.99)**
0.00001798
(19.37)**
0.24203507 0.48342151
(2.81)*
(9.39)**
Violence frequency
(1)
(2)
-0.00026 -0.00023379
(2.46)*
-2
-0.35696863
-2.15
0.00006316
-0.82
-0.00000576
-0.37
0.3108 0.38118616
(7.16)**
(4.83)**
Table 10: Reduced Form Estimation by Crop State - Violence Frequency
Observations
13751
13751
5323
5323
Colmuns (1) and (3) : Controls - rainfall shock
Columns (2) and (4): Controls - all excluded as well as included instruments other than the
total household income
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Constant
Distance
Arable land (per capita)
Total crop area (per capita)
Rainfall shock
Wheat major state
Rice major state
Reduced-form
36
Table 11: Reduced Form Estimation -Violence Incidence and Frequency - Pulled
19074
Violence frequency
(3)
(4)
-0.00029 -0.000306
-1.85
-1.9
-0.000109
-0.32
-0.481074
(2.16)*
0.000741
-0.89
0.055131
-1.43
-1.22E-05
(6.47)**
-4.14E-06
-0.28
0.375648 0.388742
(4.02)**
(3.81)**
Observations
19074
19074
19074
Colmuns (1) and (3) : Controls - rainfall shock
Columns (2) and (4): Controls - all excluded as well as included instruments
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Constant
Distance
Arable land (per capita)
Rice major state
Total crop area (per capita)×shock
Total crop area (per capita)
Rainfall shock
Rice major state×shock
Violence incidence
(1)
(2)
-0.000137 -0.000143
-1.8
-1.85
-8.37E-05
-0.51
-0.229583
(2.16)*
0.000461
-1.18
0.031952
-1.71
-6.07E-06
(8.36)**
9.8E-07
-0.16
0.188137 0.193119
(4.26)**
(4.04)**
Reduced-form
37
19074
Observations
19074
Column (1) and (3) does not include household labor income
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
19068
19068
13.7
Income contribution
(3)
(4)
0.00455519 0.00450643
(3.59)**
(3.51)**
-0.004998
(2.79)*
1.65592424
(2.35)*
0.00610613
-1.78
0.20833838
-0.98
0.0000357
(5.30)**
0.00002578
-0.94
-0.432348 -0.3927611
-1.6
-1.42
Table 12: First Stage - Working Status and Income Contribution
31.75
F-test
Constant
Distance
Arable land (per capita)
Rice major state
Total crop area (per capita)×shock
Total crop area (per capita)
Rainfall shock
Rice major state×shock
Working status
(1)
(2)
0.00195794
0.00194796
(4.04)**
(4.00)**
-0.00213503
(3.86)**
0.87813451
(2.58)*
0.00211202
(2.30)*
0.03707016
-0.35
0.00001747
(4.37)**
-0.00000552
-1.32
-0.14078789
-0.13268264
-1.15
-1.07
First stage
38
0.10432387 0.17718824
(6.65)**
(3.74)**
-0.0574957
(5.70)**
(1)
(2)
0.06018221 0.04610125
(5.70)**
(4.42)**
0.223101
(4.90)**
(3)
0.04467477
(4.61)**
-0.00137301
-1.32
-0.00048979
-0.63
0.0059713
(2.35)*
-0.00322918
(2.64)*
-0.00069247
-0.68
0.0199553
-2
-0.04020909
(4.72)**
0.17831943
(3.91)**
(4)
-0.06973191
-1.62
-0.00035761
-0.32
-0.00060421
-0.76
0.00422669
-1.56
-0.00390892
(3.33)**
-0.00193528
-1.75
0.03602574
(2.51)*
-0.05353467
(6.07)**
IV
0.31
(5)
-0.07467334
(1.96)*
-0.00075896
-0.69
0.00001157
-0.02
0.0032852
-1.07
-0.00342731
(2.69)**
-0.00201326
-1.63
0.0409737
(2.42)*
-0.06076637
(4.89)**
-0.00075896
-0.69
0.13917049
(6.36)**
Table 13: Women’s Working Status and Incidence of Spousal Violence
Observations
19074
19074
19074
19074
19074
Column (4): rice major state × rainfall shock is the only instrument
Column (5): rice major state × rainfall shock, rainfall shock, total crop area per capita, total crop area × shock
rice state dummy, per capita arable land, distance from the nearest town are instruments.
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Overidentification (Chi sq p-value)
Constant
Household labor income
Wealth
Low caste
Man education
Woman education
Number of children
Man age
Woman age
Working status
OLS
Violence incidence
39
0.10432387
(6.65)**
(1)
0.12530626
(5.15)**
0.17718824
(3.74)**
-0.05749572
(5.70)**
(2)
0.0945618
(3.91)**
0.223101
(4.90)**
(3)
0.08977954
(3.78)**
-0.00309786
-1.38
-0.00015142
-0.09
0.0093499
-1.64
-0.00608519
(2.39)*
-0.0027859
-1.47
0.03964979
-2.07
-0.04020909
(4.72)**
IV
0.26252133
(3.12)**
(4)
-0.31814208
-1.69
0.00039661
-0.14
-0.00049797
-0.28
0.00324039
-0.44
-0.00901641
(2.91)**
-0.00738072
(2.93)**
0.09849119
(2.53)*
-0.13583885
(4.87)**
0.33
(5)
-0.21203088
(2.45)*
-0.00132974
-0.56
0.00082313
-0.53
0.00299866
-0.47
-0.00673927
(2.47)*
-0.00611696
(2.57)*
0.09117923
(2.57)*
-0.14120235
(5.42)**
-0.00179615
(2.00)*
0.27467359
(5.51)**
Table 14: Women’s Working Status and Frequency of Violence
Observations
19074
19074
19074
19074
19074
Column (4): rice major state × rainfall shock is the only instrument
Column (5): rice major state × rainfall shock, rainfall shock, total crop area per capita, total crop area × shock
rice state dummy, per capita arable land, distance from the nearest town are instruments.
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Overidentification (Chi-sq p-value)
Constant
Household labor income
Wealth
Low caste
Man education
Woman education
Number of children
Man age
Woman age
Working status
OLS
Violence frequency
40
0.11093947
(6.81)**
(1)
0.01854661
(7.36)**
0.17493061
(3.71)**
0.21905421
(4.89)**
(3)
0.01309933
(4.47)**
-0.00126946
-1.22
-0.00051771
-0.65
0.0058194
(2.31)*
-0.00333443
(2.72)*
-0.00072206
-0.74
0.02082403
-1.94
-0.05931248 -0.04164783
(5.60)**
(4.73)**
(2)
0.01367613
(4.42)**
(4)
-0.02985857
-1.53
-0.00035413
-0.31
-0.00052948
-0.69
0.00411542
-1.52
-0.00386254
(3.30)**
-0.00221614
-1.65
0.03844753
(2.50)*
-0.05410003
(6.28)**
-0.00035413
-0.31
0.17544249
(3.92)**
IV
0.3
(5)
-0.02496323
-1.55
-0.05985892
(5.18)**
-0.00089894
-0.85
0.00011331
-0.17
0.00339135
-1.14
-0.00317429
(2.37)*
-0.00209937
-1.61
0.04014476
(2.41)*
-0.0007452
-1.87
0.13457674
(6.18)**
Table 15: Women’s Income Contribution and Incidence of Spousal Violence
Observations
19068
19068
19068
19068
19068
Column (4): rice major state × rainfall shock is the only instrument
Column (5): rice major state × rainfall shock, rainfall shock, total crop area per capita, total crop area × shock
rice state dummy, per capita arable land, distance from the nearest town are instruments.
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Overidentification (Chi-sq p-value)
Constant
Household labor income
Wealth
Low caste
Man education
Woman education
Number of children
Man age
Woman age
Income contribution
OLS
Violence incidence
41
0.21726551
(6.56)**
(1)
0.04057954
(6.12)**
0.35699338
(3.73)**
0.44200342
(4.59)**
(3)
0.02848227
(3.96)**
-0.00293735
-1.33
-0.00020573
-0.12
0.00913164
-1.66
-0.00626973
(2.46)*
-0.00277106
-1.55
0.04051786
-1.97
-0.13192466 -0.09452102
(5.91)**
(5.01)**
(2)
0.03015058
(3.96)**
0.34
(4)
(5)
-0.06345346 -0.06408688
-1.6
-1.86
-0.00097842 -0.00181917
-0.41
-0.8
-0.00023093
0.00105567
-0.14
-0.72
0.0054849
0.00364871
-0.96
-0.6
-0.00739996 -0.00597215
(3.09)**
(2.22)*
-0.00596859 -0.00609695
(2.29)*
(2.28)*
0.07823453
0.08590052
(2.63)**
(2.65)**
-0.1211704 -0.13672303
(6.53)**
(5.80)**
-0.00188798
(2.13)*
0.34866843
0.26012387
(3.73)**
(5.29)**
IV
Table 16: Women’s Income Contribution and Frequency of Spousal Violence
Observations
19068
19068
19068
19068
19068
Column (4): rice major state × rainfall shock is the only instrument
Column (5): rice major state × rainfall shock, rainfall shock, total crop area per capita, total crop area × shock
rice state dummy, per capita arable land, distance from the nearest town are instruments.
Robust t statistics in parentheses
* significant at 5%; ** significant at 1%
Overidentification (Chi-sq p-value)
Constant
Household labor income
Wealth
Low caste
Man education
Woman education
Number of children
Man age
Woman age
Income contribution
OLS
Violence frequency